The JGroups team recommend the following tuning for the transport:

thread_pool.enabled="true"
thread_pool.queue_enabled="false"
thread_pool.max_threads=<available-cpu*scale-factor>